STAGE-Claw: leaderboard

Metric: Average task score (0-100) from the verification programs, 40 automatically built and human-audited personal-agent tasks run in real operating-system environments (calendar, email, reminders, files); the final system state is checked by task-specific verification programs; OpenClaw agent framework, reasoning disabled where the model allows it, temperature 0 where configurable, first valid run per task; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 11 models tracked.

Top models

#ModelScore
1Claude Opus 4.777.1
2Claude Sonnet 4.669.43
3GPT-5.5 (Non-reasoning)69.19
4GLM-5 (Non-reasoning)67.4
5Gemini 3.1 Pro (Preview)65.5
6GPT-5.4 (Non-reasoning)65.37
7Seed 2.0 Pro (Non-reasoning)61.07
8DeepSeek V4 Pro (Non-reasoning)59.78
9Qwen 3.5 Plus (Non-reasoning)59.74
10Kimi K2.6 (Non-reasoning)59.05
11MiniMax-M2.747.53

Interactive version: theaggregate.ai/benchmark?slug=stage-claw · How It Works · Data refreshed daily, snapshot 2026-09-29.